Noting that Petite has a very well integrated and optimized expander, while Alexpander being the most general/portable one. So this is a very good result.
Also note that macro-computation, having a normal-order, head-first term-rewriting system at its core is in fact very close to a Turing machine.
Computing (is-prime 7) at compile-time (on an old 1.7Ghz laptop with 2G memory) gave the following (very rough results):
- Petite (7.4d): 1.4s
- Alexpander/Stalin (0.11): 4.8s
- Alexpander/Bigloo (3.6a): 8.5s
- Chicken interpreter (4.2.0): 11s
- SCM (5e5): 15s
- Racket (v5.0.1): 39s
- Ypsilon (0.9.6-update3): 47s (fails on larger primes)
- Larceny (0.97): 54s
- Plt/MzScheme: (v4.2.1) 110-121s
- Mosh (0.2.6): crash after XXX minutes
- MIT-Scheme: crash after XXX minutes
- Scheme48 (1.8): crash after XXX minutes
- Scsh (0.6.7): FAILED
- Gauche: FAILED
- Gambit: FAILED
- Ikarus: FAILED