I have no idea if it is still relevant, but here are some time measurements from my 2 PCs (both latest beta 36 obviously).
I tried to pick a value that seems to be about average for a command (only integer values for larger numbers as the decimal places don't carry any actual meaning)
Values in parenthesis are the approx. deviation (also estimated by me from results I've seen, not calculated) over multiple executions of the same command, so the real values are in the 'Value +/- deviation' range.
Laptop running Win7 64-bit (Aero off), Core2Solo U3500 @ 1.4 GHz
alpha 0: 9.780 ms (~ 100 ms)
alpha 1: 9.720 ms (~ 150 ms)
alpha 2: 9.985 ms (~ 150 ms)
alpha 3: 10,825 ms (0,2 ms)
alpha 4: 30,358 ms (0,5 ms)
Main PC running WinXP 32-bit, Athlon 64 X2 5600+ @ 2.9 GHz
alpha 0: 135 ms (25 ms)*
alpha 1: 135 ms (25 ms)*
alpha 2: 135 ms (25 ms)*
alpha 3: 15 ms (2 ms)
alpha 4: 25 ms (4 ms)
*) most values were in the given range, sometimes I got values of: 35 ms (3 ms)
this was mostly the case when I was hammering the enter button (with the command in the address bar), i.e. many executions in a short time.
due to the high deviation alpha 0 through 2 are basically indistinguishable on my main PC, can't really tell which is faster (if any)
My Conclusion:
Even assuming the shorter range of alpha 0 through 2 on my main pc are the values relevant for practical applications, alpha 3 seems to be the fastest variant in all cases... At least judging from these tests I don't see any reason for special code for Win7 (I assume there currently isn't any?). So this problem of Win7 seems to have led to the implementation of an overall much faster drawing routine? nice
