A good way to think of it (not necessarily correct, but I think it's representative in this case): there is leakage in both the C-B and B-E junctions.
When testing Vceo, the base is open circuit, so any leakage current flowing from collector to base causes forward-bias in the base, which turns the transistor on slightly. That turning-on current flows out the emitter, which is fine, but the base kind of gets dragged along for the ride. The overall effect is, the transistor acts like a zener diode around 30V or so.
If, instead of allowing the base to float, that leakage current is drawn from it directly, then there is no gain effect, and the breakdown voltage is as high as possible. This is Vcbo.
And yes, one can effectively use a transistor as a variable zener diode, controlled by placing a resistor B-E to shunt away some of that leakage current. (I doubt the voltage is at all stable, let alone
noise free. So don't actually use this for something, just be aware that it exists.)
The same effect should occur on the E-B breakdown, i.e., Vebo > Veco. Come to think of it, I haven't tried this, I should measure it.
The practical effect of this is: if you're in a switching circuit... it depends! If your circuit actively shunts the base to ground with a seriously strong driver, then it will certainly be Vcbo*. If it's a much weaker drive, like a simple B-E resistor, it will probably be at Vceo for some time.
*Except when it's not. It also takes some time to get there, too. If the transistor was just turned on hard, it will act like it's still slightly on for some time as excess charges dissipate. So a transistor might avalanche at Vceo while it's turning off, then gradually rise to Vcbo. This is important on high voltage switching transistors, like MJE13007, or horizontal output types. These often specify the RBSOA (reverse bias SOA), which have curves showing how much voltage you can expect to withstand after however much time.
Practically, I would treat Vceo as equivalent to a MOSFET's Vdss. That is, a 40V transistor will be okay for switching no more than 25 or 30V -- allowing for supply variation, supply transients, and the normal switching transients that occur on every cycle.
By the way... experimentally, it may be significantly higher than rated, too. I typically measure 2N3904s breaking down around 100V.
Tim