This is bug report for a new sample file, attached to a unrelated bug 702381 as the comment #8. Ghostscript takes extremely long time to parse a CMap with 38917 entries in a single beginbfchar ... endbfchar range. This inefficiency is caused by repeated use of --roll-- on the whole data set, which results in quadratic complexity. A simple rewrite of beginbfchar restores linear complexity and reduces parsing time to 1 sec. The function endbfrange has the same problem, but it is not changed, since it is not used by the sample file.
Fixed in commit fd919a2ed. Thanks, Peter.