why regexprep much slower than strrep

Consider the following simplified example:

str = cellstr(repmat('987_890',1e5,1));
timeit(@()strrep(str,'_',''))
timeit(@()regexprep(str,'_',''))

strrep is ~3x faster than regexprep() on R014b - Win7 - cpu i5 3rd gen.

I reckon regexprep carries some overhead for being more flexible but I would expect it to be slightly slower than strrep unless it does the replacement on the go (to allow backtracking). However, in this simple substitution I hoped the engine was smart enough to recognize it does not need to backtrack.

Any thoughts?