for
String s1, s2;
s1 += s2;

we can generate slightly better code by using

s1 = s1.append(s2);

which allocates twice: a String and a byte[] once, compared to what we currently generate

new StringBuffer().append(s1).append(s2).toString();

which allocates at least thrice: a StringBuffer, a byte[], a String, and possible more byte[] arrays during buffer expansion. An intermediary would be

new StringBuffer(s1).append(s2).toString();

which would at least avoid expansion for the first append.

Apparently the old compiler performed this optimization.