Hmm, wonder if that was gold doped. Would explain the higher leakage (arguably; leakage at 25°C is low enough it might be a testing limit, but it's also not a very round number?), and the relatively low voltage and switching time. hFE curves are not provided (steeply reduced hFE at low Ic is a hallmark of such types).
Higher C-B leakage I think is a factor in Vcbo vs. Vceo breakdown, since the leakage is effectively multiplied; I'm not sure what all physics is really at work, as the ratio can be quite significant (as in this case); other times it's much more modest.
The boundary between the two conditions is a kind of avalanche breakdown that generates very high noise, indeed the avalanche multiplication can extend so far that the device saturates in a fraction of a nanosecond, while carrying several amperes in the process. (Evidently this phenomenon involves two mechanisms: a high-level injection mechanism which crowds out the base doping concentration, effectively causing local punch-through (effectively a C-E short); and subsequent rapid heating, causing formation of a current filament (at high temperatures, intrinsic carriers outnumber dopants, maintaining punch-through). That perhaps explains why recovery from this mechanism is so slow -- several microseconds at least.)
FWIW, I'd love to see microscopy of a bare die under such operation -- but I'm also afraid that there would be so little light emitted (due to the very low duty cycle, < 0.1% is typical) that it might not show at all.
(The usual setup is: a large pullup resistor to charge the collector, typically 10-100k to +100V or thereabouts; a B-E resistor, typically 4.7k or thereabouts, depending on type; and some kind of load, at least a few pF from C-E but also including an output coupling network like a 50 ohm transmission line or whatever. 100-120V is enough to cause 2N3904 to break down in this way; 2N2369 needs less, 60-80V I think; it even works for high voltage types, but because breakdown occurs as a narrow filament, power transistors are essentially dead weight for this -- a 1500V 10A transistor fails with only a few 100s of pF load, hardly more than a 300V 100mA type can handle.)
Tim