Wrt the user experience (specifically, time spent in the CPU) impact
of the extra GC, I put some GetSystemTime calls around the new
CleanUpJsGlobalVariables function. On tester/gui.html, the clock time
is indistinguishable with and without the CleanUpJsGlobalVariables
patch - trying both with and without twice gives me about 5 minutes
+/- 10 seconds to go through the entire test suite, on my setup - the
difference with and without the memory leak plugged is
indistinguishable from noise. The deltas in the GetSystemTime calls
are also zero, i.e. less than the granularity of GetSystemTime. I'll
take the zeroes with a grain of salt - this is timed from a WinXP
running as a VMWare guest of a multi-core workstation, and I have had
weird clock issues before, on this setup. But, still, I think that the
performance impact of this memory leak fix is negligible.
