Looking back at
SOLR-2977 and r1222234 where these lines of code were added, i think it was just a brain fart on grant's part – he introduced a code branch in a place where (previously) BooleanClauses were always added with MUST_NOT and now that he didn't wnat to actaully exclude them, just mark them, he cloned that code and had it use SHOULD instead.
but i agree, that doesn't make any sense – the set of results using markExclude=true should be the same as the set of results you get not using the elevation component at all, and the scores of the excluded docs shouldn't change.
Committed revision 1384567.
Committed revision 1384577. - 4x
Thanks for finding this and providing the test alexey!