Hi
>If a query with a small result set is not based on a good selectivity
but a full table scan and a grouping operation , it cannot be fast, IMO.

If that is the case, I'ld opt for the xBase - solution.
This would get [relatively] better the larger the rowsize of
b.cfk1, b.cfk2, a.ctype
really is, since no temp table to later group on is needed.

Simulating group by in xBase is not that hard <g>,
and even scanning the whole "many" shouldn't take more than 10 secs,
if the rowsize/hardware of the "many" allows fast throughput.
Processor speed nearly irrelevant here, I/O bottleneck IMHO.

But without better info this is just shooting in the dark.

my 0.02 EUR

thomas