Good timing Scheme is undergoing quite a bit of change that should culminate in about 6 months.
I have used scheme at work and for the last 6 weeks on a private project in earnest.
One issues is that there are a lot of scheme implementations out there. Some good, great, average, and others bit rotted. Only mention two here.
DrScheme/MzScheme is pretty amazing. It is fully loaded and few know of its full capabilities. The fact that PG can just "write" Arc in MzScheme and some other pretty neat things that are doable in MzScheme is not an accident. DrScheme/MzScheme is much, much more then Scheme in a way not too many people understand.
The second implementation is Larceny. True native compiler. Plenty fast. The compiler is written in Scheme (small C runtime). Its hackable from top to bottom. Extensible and did I say fast? Threading is fair or green threads using continuations/call-cc (no O/S threads). Don't like the Scheduler? Modify it, its in tasking.sch The interrupt handler is a lambda function, just redefine it. Don't like the i386 code output, hack the assembler or the optimized 5 pass compiler. Its all in Scheme.
Don't like standard poll, I added epoll via the FFI in a few hours.
Larceny has been around for years, but is undergoing a resurgence to bring it up "closer" to the new R6RS standard.
I think it has great potential. Be advised that do to the rapid changes in development underway that there are some rough edges in certain "library" areas, however, the core compiler is solid and stable.
Awhile back Slime was ported over to Scheme48 (not a system for general real world use, but a fun hacking implementation). The compiler group should have Larceny stable with many of the libraries and features of the new R6RS standard (some will not implemented). IF SLIME scheme backend was ported over to Larceny as well. It could be the start of something special.