"word-break" is new property of CSS3 text module. First implementation is Microsoft's Browswer as known as Internet Explorer as their origianl specification.
Internet Explorer 7 and Safari 3.1 has word-break implementation. So I try to confirming the "word-break" implementation on both browsers.
Safari 3 / Google Chome (WebKit)

WebKit doesn't consider CJK scripts. The result of all modes is same. If using "word-break: keep-all;", KANJI characters should not break line.
Internet Explorer 7

Trident's behaivour of Japanese scripts is strenge. When using "word-break: keep-all;", although "TOUTEN" ("、". It means like comma) is broken lines, KUTEN ("。". It means like priod) isn't broken lines.
No one has correct implementation of "word-break" property.