I most definitely prefer the quartz method of resizing windows, much better than XP.
As for the speed, the reason is three fold, and this is based off my research, so it may not be 100% correct (but I'm a developer, so at least I know most of it is).
- GCC 3.3 - It has a new method of arranging the instructions for a PPC processor (G3, G4, G5) which has brought about a massive performance boost (bigger than from 2.9.5 to 3.1 in my testing). This should only become better as time progresses and Apple keeps adding optimizations to the compiler.
- Quartz - From my research Apple spent a considerable amount (read 20%) of all the OS teams resources on making quartz faster and less of a memory hog. They have improved the window buffering time, the drawing to the screen time and lowered the GPU requirements when handling windows. Window resizing is faster (most definitely in the Finder) but not 'that' fast as to write home about it. Its most definitely an improvement over 10.1 speeds, but not as much as 10.2. (This is from testing on a GeForce 4MX and GeForce 4 Go). Quartz also seems to handle separate drawing threads better, in that a user interface element in Jaguar could drag down the entire interface if it was needing to be redrawn a lot, now that seems to be improved and negated.
- Memory Management - As far as I can tell, the handling of the cache and RAM has improved drastically. Mac OS X now keeps a handy 12-20mb of RAM free for instantaneous tasks when its nearing the max amount of RAM which has the advantage of improving the perceived performance of the interface markeably as the machine gets bogged down. Also, applications that are not being used are handled much better in the cache allowing the applications you are using to perform to their full potential. Overall everything is snappier from these improvements.
I think this is a good example as to what all these changes have done...on my PowerBook 867, iTunes 4 in Jaguar uses are 12-14% CPU based on my normal work usage (EQ and sound enhancement on). In Panther, it uses 7-9%, a drastic increase in performance with little to no code changes.
I hope this assists everyone.
Mat