Because it is not capable of ordered index scan that handles both filtering (by multiple conditions) and "order by" - text search returns potentially a very large set of results that must be sorted and then truncated (to satisfy limit).
EDIT: or I misunderstood the question. GIST+trgm works but its search capabilities are limited, it does not handle "order by column" so requires weird workarounds, and it is not very efficient.
EDIT: or I misunderstood the question. GIST+trgm works but its search capabilities are limited, it does not handle "order by column" so requires weird workarounds, and it is not very efficient.