Note that this happens with multiplication also. There's a classic study of anchoring bias where people are asked to estimate about how large 1*2*3*4*5*6*7*8 is or are asked to estimate how large 8*7*6*5*4*3*2*1. People asked either underestimate on average and underestimate the first much more than the second.

This makes me suspect that there's a general tendency to underestimate independent of what operations are being performed.

ETA: Some of this is discussed (with citations) in a previous post by Eliezer here.