So I thought I’d try out Kjartan’s performance-enhanced vte test tarball, and got some interesting results from a few entirely unscientific tests. find / on my laptop provided a 25M file, which I exposed to the terminal emulator via time cat.
For Kjartan’s preformance-enhanced vte code:
real 0m29.826s user 0m0.002s sys 0m2.871s
For xterm with default settings (bitmap font):
real 0m52.172s user 0m0.004s sys 0m2.712s
For xterm with the same font, using Xft:
real 6m3.874s user 0m0.003s sys 0m0.484s
And finally rxvt with default settings (bitmap font):
real 0m16.845s user 0m0.009s sys 0m4.813s
Now, this is an entirely brutish test, because very few use cases involve non CPU intensive work that requires so much rendering effort from the terminal. It would be interesting to do a kernel build comparison, to see how much CPU the terminal chugs during a real world task.
Andrew Bennetts points out that vte cheats by not displaying all the output (which is fairly reasonable, in my opinion), and that it appears slower due to intermittent, halting, half-second delays - a smoother appearance should feel faster.
Meanwhile, yeeeeee-haaaah! Thanks to Kjartan for pulling these changes in! 
Update: Two co-workers, Colin Watson and Daniel Silverstone, encouraged me to try the same test with pterm, which is based on the terminal code that powers PuTTY, and uses GTK+ 1.2. I’m told that patches to port it to GTK+ 2.0 would be (relatively) warmly accepted.
real 0m13.487s user 0m0.005s sys 0m2.795s
Dave Neary and aforementioned co-workers also pointed out that I didn’t provide numbers for the current release of vte, without Kjartan’s performance-enhancing patch collection. That was pretty stupid. So, here it is:
real 2m26.535s user 0m0.003s sys 0m2.609s
