Memory bench as accurate speed test?

[SuperPickle]

I recently did a battery of memory benchmarks with Sandra at different multiplier/fsb settings all at default voltage. I found my bandwidth to be higher at 9.5X147 than 10.5X143 even though the processor speed was 100MHz slower. This makes sense to me as the fsb is higher but my question is: Is this an accurate messurement of overall speed or are the raw clock cycles more important. Any input is appreciated. Temps differ by 3 degrees at load.

ABIT KG7-RAID
1.33 AYHJA 'Y'
512M PC2100 Crucial@CAS2
Radeon 64DDr