>>33
Well, this discussion isn't about GC vs MMM anymore
Good. That's boring.
I don't really expect GC to beat a modern malloc. I am cool with both automatic and manual methods but I'll take the least sophisticated malloc in common use (probably dlmalloc which appears in msys) or refcounting (with or without cycle detection) over a glacial GC (e.g. Java's) any day.
Today we don't have an issue with speed and we rarely have an issue with memory use. Latency is a big (but not universal) problem in GC'd languages, so I don't mind wasting more time (and space) doing GC as long as the user doesn't notice the difference. I know I pick on Java too much, but traditionally it tends toward explosive memory use, incurring a long-running collection, violating both of these conditions. I hear it's not so bad lately.