It is probably not significant since it is a perfectly predictable branch, but it should be measured nonetheless. If the cost of the comparison on every call to VERBOSE_LOG() turns out not to be negligible, the macro should probably be modified to be a no-op in release mode.
After discussing it with Saam, I eventually decided to just make VERBOSE_LOG a no_op in release mode by default, with an easy way to change it. See https://bugs.webkit.org/show_bug.cgi?id=177922 for details.