Seems like the problem shows itself when the total number of trailing decimal zeroes in all numbers of the calculation exceeds 8.

So round( .3500000000 * 943 * .1, 2 ) = 33.01
round( .3500000000 * 943.0 * .1, 2 ) = 33.00
round( .350000000 * 943.0 * .1, 2 ) = 33.01
round( .35 * 943.00000000 * .1 ) = 33.01
round( .35 * 943.00000000 * .10 ) = 33.00

Playing around further while writing this, I've found the following :

.35 * 943 * .1 = 33.005
.350 * 943 * .1 = 33.0050
.3500 * 943 * .1 = 33.00500

It seems that the precision of the result of the calculation is based on the precision of the operands. It maybe that the value transferred to the function may not retain the required precision.

Another one to try is
.350000000000 * 943 * .2 = 66.0100000000000
.3500000000000 * 943 * .2 = 66.0099999999999

The extra information may not be of any help as to why the problem occurs, but it may help to avoid it.
Paul