Unity has three huge problems. The first is that they gratuitously seal classes in their API. This means you cannto override their functionality if you don't want to do things "thier way." Secondly, they dont expose anything but the top most layers of their API. (Example, they expose their physics system, but not the PhysX API on which it is based.) This means you cannot peel off layers of their API to build your own functionality. FInally, if you want source code so you CAN do things in a way other then they tell you to, it costs "well into the 6 figures" according to the Unity sales creature I talked to.
A wise man once said that a tool makes simple things easy and complex things possible. A toy, on the other hand, makes simple things easy and complex things impossible. By this rubric, Unity is a very nice toy, but no more.
Also note that ANY technology that requries the user to install a plug-in cannot be used in schools or libraries or other public Windows system where the user does not have administrator privledges. At GDC this year Unity made noises about a version that will deploy to Flash to get around this problem BUT there are a whole lot of unanswered questiosn about this product including performnance (Flash VM sucks) and physics (there is no PhysX API for Flash.)
WebGL has one HUGE stumbling block, that is Microsoft. MSFT hates OpenGL as it competes with Direct3D. It's likely to be a cold day in hell before they support it in IE which, depending on whose statistics you want to believe, is still 30% - 60% of the browser market. Our best hope there is that MSFT releases a "WebDX" instead and some enterprising Javascript scenegraph maker builds a layer that runn on either.
O3D is dead. It's that simple. Google has no manpower on it according to the Google guys I talked to at GDC this year. Theya re-focusing on WebGL and want others to write the scenegraph. Copperlicht looks pretty nice if thaqts the direction you want to go in.
Torque3D has a reputation for beign really buggy and, again requires a plugin to run in a web browser.
Papervision's peformance up to now has been REALLY low. We're talking a handful of flat shaded ploygons and no more. The new Flash 3D acceleration might help that. Personally I also was underwhelmed by their functionality, but look at their API yourself.
So there really IS no good answer yet to your question, sorry. I'd bet my money on WebGL except for the Microsoft problem, which is a big one.
Edit: I've started looking at Away3D 4 and I am pretty impressed so far. Its a shame that thier Webgl/Javascript version is two versions back, but you could always wrap their Flash engine in a Javascript API. Something I'm actually considering doing...