Cadence APS simulates slower??

[gszczesz]

I have a digital, synthesized circuit that is embeded in an analog IC.
I wanted to compare which method was faster:
1) Spectre - replacing all MOS devices with ideal switches and capacitors
2) Turbo - replacing all MOS devices with ideal switches and capacitors
3) Turbo - Using real MOS models
4) APS - replacing all MOS devices with ideal switches and capacitors
5) APS - Using real MOS models.

Spectre and Turbo both simulated about 2-3X faster using ideal switches, as expected. The suprising thing was APS: it simulated 2-3X SLOWER than Spectre/Turbo with ideal switches. APS did perform faster with real MOS than Spectre/Turbo with ideal switches.

The timing was as follows:
1) 18m (conservative), 10m (moderate)
2) 18m (Conservative), 10m (moderate)
3) 32m (Conservative) = same as spectre with real MOS.
4) 37m !!! (Conservative)
5) 14m (Conservative)

So APS did SLOWER with ideal switches than real MOS... how can that be?

Greg