Should we distinguish between predicates that are projectable and predicates that are natural? "Green", "ultraviolety-looking to bees", "owned by Rachael", and "tasty to alpacas" are fairly projectable, but I doubt they're natural. I don't expect a bee to care which objects are green, a mouse to care which objects are owned by Rachael, or a human (who doesn't have a special interest in bees or alpacas) to care which objects are ultraviolety-looking to bees or tasty to alpacas. There's a sense in which all the above concepts are parochial, and are only worth adopting because they're convenient. Furthermore, there's a sense in which there's no conflict between these different ways of carving up the world: I don't think the alpacas are wrong, and I might find it very important to adopt their concepts when interacting with them.
Also, maybe "Having mass m before time t or mass m* after" is an example of a natural predicate that isn't projectable.
That last example sounds gerrymandered (i.e. not a natural way to carve things up) to me. I'd need to hear more about why we'd consider it 'natural'. (Note that complex predicates may invoke natural terms like 'mass' in a gerrymandered or non-natural manner.) On the other hand, I'm also unsure why green and ultraviolet are thought not to be very natural. There is a natural respect -- namely, surface reflectance properties -- in which all green things (or all ultraviolet things) exhibit a genuine similarity.
The other examples are plausibly less natural, but also (and, I imagine, to the same degree) unprojectible. Here I should clarify something that may not have been clear in my first post. 'Projectibility' does not just mean stability across time, i.e. that if something satisfies the predicate now, it will also do so in future. Any tenseless predicate, however gerrymandered -- e.g. 'exists in 1907 or 2008' -- will be temporally stable in this sense. Genuine projectibility is more general, since we can inductively project along dimensions other than time. If a bunch of Fs are green, that may be evidence that other Fs (and not just these same ones in future) are also green. On the other hand, if a bunch of Fs are owned by Rachael, that's probably not enough to justifably project ownership-by-Rachael onto any other arbitrary F.*
Overall, then, I remain of the opinion that 'natural' predicates, i.e. those that carve nature at the joints, or highlight objective similarities, are also those that will tend to support inductive projection.
[* = I'm not sure if that's the best way to explain projectibility. Any suggestions?]