After restoring from my 10.5.5 backup drive (as mentioned above, the 'partial' 10.5.6 Apple Software Update messed up my machine as opening DMG files for one thing caused instant kernel panics), I reinstalled 10.5.6 using the combo update and all appears well now.
I read that 10.5.6 has some graphic/gaming improvements, so I decided to run Xbench again to compare (not the greatest tool overall, but OK for comparing against earlier results against itself within some constraints at least; it does seem to vary too much over several tests, IMO though).
Here's the overall results compared to 10.5.4 and 10.4.11 on a 1.8GHz G4 with 1.5GB of L2 latency ram and an ATI 9800 Pro and Sonnet SATA.
-CPU score is noticeably faster under 10.5.6 (84.3) than 10.5.4 (71.1), but still slightly lags Tiger (88.06). Good to see Leopard catching up, though.
-Thread test up slightly in 10.5.6 (65.8) versus 10.5.4 (62.5) but still lags Tiger significantly (79.7)
-Memory test is up in 10.5.6 (39.1) versus 10.5.4 (36.9) which was already slightly faster than Tiger (36.1)
-Quartz graphics is down (89.5) versus 10.5.4 (93.6) but still higher than Tiger's (85.2). Text rendering took the biggest hit in 10.5.6, dropping from 256 in 10.5.4 to 152 in 10.5.6! That is still faster than Tiger's 128 score, though. Still, I wonder what happened there as it's almost half 10.5.4's score (unless Xbench is that unstable?) Edit: I would say it's Xbench alright. Retests gave text scores anywhere from 100 to 350!
-OpenGL tests are higher in 10.5.6 (79.6) versus 10.5.4 (73.0) but still pale next to Tiger's score (90.9)
-User Interface Test is down in 10.5.6 (48.5) from 10.5.4 (50.6) and (83.1) in Tiger. It's sad to see that Leopard still lags in this area quite a lot from Tiger. The 9800 Pro has full Core Video and Core 3D support in Leopard so it's not that.
-The disk test seems to vary a lot from test to test so I don't want to jump to really big conclusions, but I do find it interesting where the changes are at. Overall, 10.5.6 scored lower (68.3) than 10.5.4 (75.4), but similar to Tiger (68.0). However, uncached write is nearly double of Tiger's score in 10.5.6 (88 versus 44.5) but 10.5.4 had a score of 115 there. OTOH, I've seen Tiger's disk score reach almost 100 before the last video card and USB card upgrades so there's some odd variances there).
Overall, 10.5.6 gained some significant ground over 10.5.4 (CPU, Memory, Thread, OpenGL), but also lost some minor ground in other areas (Quartz, User Interface and Disk with a huge possible loss in text rendering, although still better than Tiger (again I reiterate Xbench isn't that great of a consistent testing tool, but then what else is there?).
Overall scores were better for 10.5.6 (62.7) versus 10.5.4 (61.1) but still lagged Tiger (69.2).
Hopefully, Leopard will continue to be tweaked. With the rumor that Snow Leopard will dump PPC support, I'm hoping they can at least get Leopard up to Tiger speeds before they're done with it. The differences are small enough now and the bugs addressed well enough, though that I feel I could use Leopard on my PPC machine on a daily basis except that I also lose Classic mode. I do not use many classic apps these days, but it's nice to have available and a bit of a shame they could not have carried it over as an optional install for PPC machines as it's the main reason I still run Tiger (the only app I have that is Leopard only is Handbrake and I tend to run it on my MBP for the speed differences anyway).
Edit:
I ran XBench again under 10.5.6 and to give you an idea of how unreliable Xbench is here are some notable differences just between one test to the next (first number is 2nd test). I'm only posting significant differences; others were very similar numbers:
Overall: 63.0 versus 62.7
Cpu: 91.6 versus 84.3
Quartz: 93.2 versus 89.5
Disk: 65.1 versus 68.3
A 3rd test (shutting down background tasks I normally run) erased 10.5.4 differences in several areas where earlier tests indicated a loss so I'm guessing 10.5.6 differences are not very significant or at least untestable using Xbench in these areas for reliable differences:
Overall: 63.2 versus 61.1
Quartz: 94.2 versus 93.6 in 10.5.4 (passed it where it lost before)
User Interface: 49.9 versus 50.6 (closer than before)
I've also found the "100" defaults a bit odd compared to actual system tests. For example, my disk scores are often as high as 150 on the uncached 256k test yet as low as 20 on the uncached 4k write test. That's a huge difference in performance compared to a 100/100 norm. Maybe it's the hard drive?
Overall, I think the most surprising aspect was when I ran Xbench on my MBP I just bought in October (the pre-uni-body model with a matte screen and two 2.4GHz Core 2 Duo CPUs and 2GB of ram). For all the talk about how much faster the Intel Macs are compared to PowerPC ones, let alone G4 PowerPC models, I was surprised to find that the overall score was only about 2x faster than my upgraded PowerMac. User Interface was the biggest difference at about 6x faster. OpenGL was a mere 2x faster. The CPU score for having 2 cores and being Intel 2.4GHz was only 2x faster than my 7448 single core 1.8GHz G4. The thread test was 5-6x faster, though so clearly the core differences showed up there. Disk tests were noticeably slower than my PowerMac (5600rpm laptop drive versus my full size 7800 rpm drive), but still, that's pretty good for a machine originally from the turn of the century, IMO. Overall, the enhanced PowerMac isn't too shabby for such an old machine. If I had gotten the dual CPU model, I think it would closed the gap on the 6x scores to also only 2-3x faster maximum, which for being originally nearly 8 years old, isn't the night and day differences I would have expected. The machine clearly has plenty of life left in it for every day applications. I would imagine Mac Pro 4-core G5s would be VERY close to current Mac Pros, let alone iMacs, etc. It's a shame Snow Leopard wants to ditch them so hastily.
