This is to do with the way text is rendered in WebGL mode. Since WebGL provides no features to render text, a text object does the following:
1. Create a normal 2d canvas.
2. Render text with the 2d canvas (since that is a capability canvas2d has, but not webgl).
3. Create a WebGL texture the size of the object.
4. Paste the 2d canvas in to the texture.
5. Render the texture like a sprite.
This is a pretty well optimised process - it only does that when the text changes or its size changes, and the rest of the time it's basically a Sprite. Normally it's also only operating on small rectangles.
The problem is when you zoom in it increases the detail of the text by creating a larger texture - the size of the object in screen space and rendering larger text (so as large as the object would appear if your monitor were large enough to display it all). If you keep zooming in, you end up creating huge textures. Worse if you are zooming in a bit further every tick, every frame it will run the whole process again, including creating a new texture.
So this is by-design for high-detail text. The best thing to do would be to use SpriteFonts, or, er, not zoom in on text that far. I guess I could put in a max detail level clamp though so it doesn't explode. I'll leave this bug open for that.