Well, it doesn't matter at this point.  I wrote a bc script to search
for optimal multipliers (with a few parameters: no more than 2
multiplication stages, and no more than 3 bits per multiplier).  The
ones it came up with, for the hz values we (now) support, are:

hz	shifthz	mult1		mult2		error (ppm)
60	64	1.000000000001	1.00010001	.059605
60	64	1.00000001	1.0001		15.258789	<---
60	64	1		1.00010001	244.140625
96	128	1.000001000001	1.0101		3.814697
96	128	1.000001	1.0101		244.140625	<---
100	128	.1111111111	1.01001		.95367431640625	<---
1000	1024	1.00000001001	1.00000101	11.339783
1000	1024	1.0000001	1.00000100001	50.455331	<---
1000	1024	1.0000001	1.000001	430.107117
1200	2048	1.0010000001	1.100001	63.538551	<---

As you can see, the results for the 100Hz case (which is what PCs use)
are quite good!  1200/2048 is a bit too far to stretch, though.  (It
might work better if I multiply downward instead of upward.  I think
that's okay, so I guess I'll try it.)