As per the comment in Privoxy’s ticket #2530592, I’ve set up a Privoxy rule to interpret dot-less domains as search terms. This emulates the search function built into Ephy, which breaks with Privoxy’s DNS error messages. In /etc/privoxy/user.action I now have:
# Redirect dotless hostnames as search parameters to Scroogle {+redirect{s@^http://([^\\.]*)/$@https://ssl.scroogle.org/cgi-bin/nbbw.cgi?Gw=$1@}}
A fix for Scandinavian letters (note the single dot on the last line):
# Escape dotless hostnames with percent-encoding{+redirect{s@å$@%C3%A5@}}
{+redirect{s@Å$@%C3%85@}}
{+redirect{s@ä$@%C3%A4@}}
{+redirect{s@Ä$@%C3%84@}}
{+redirect{s@ö$@%C3%B6@}}
{+redirect{s@ö$@%C3%96@}}
.
The Hell? All of a sudden, none of the above is needed anymore. It Just Works.
OK, scratch that. Multi-word input works without Privoxy interfering, whereas single-word input doesn’t, and can’t be percent-encoded, apparently because the config syntax breaks Perl’s \×{xxxx} UNICODE syntax (brackets trigger an ‘invalid line’ error from Privoxy).