I'm no expert on lily's beaming system.
However, in your second example you don't break an existing beam with a bar/line
break, so it's rather different from the first where the "correct" beaming was
broken.
Not knowing anything about how Lily works, I'm inclined to agree. in 3/4 (at least here) a measure comprising only 8th-notes will be beamed straight through, thus (pseudo code):
e8[ e c' c c c c]
Whereas a 4/4 bar is beamed in two groups of four.
So you're correct, that there's something going on with default beaming being broken up. In fact, with the following MWE (also in 4/4):
\version "2.19.80"
\relative c'' {
c e, g8 a
\bar "" \break
g e g16 a b8
}
the "g8 a" at the end of the first line is *also* broken into two unbeamed 8th-notes, but the two that follow the break do not. Why would this be?
Cheers,
A