You mean the series resistors? Nah, the level shifter circuit is full CMOS -- resistors would cost supply consumption, which would be specified.
Relying on input protection diodes is a bad idea for three reasons:
1. They're intended for transient use only. This isn't always relevant, but sometimes it is.
2. The charge injection can disturb logic states, or affect propagation time, etc.
3. The injected charge returns through the substrate and nearby ESD diodes, depending on which ones are active. This increases supply consumption considerably, and can interfere with signals on nearby pins.
I've written a model for this before. It was the CD4001, including measured hFE of the "ESD diodes" that are actually BJTs. The input-to-VDD diodes actually shunt about half of the positive ESD current to ground (the current is cascoded through a PNP with low hFE). A small fraction (about 3%) leaks out adjacent input pins if applicable (ESD diodes are typically per-gate, so in a 4001, this causes leakage out the other input pin, but doesn't affect other gates; this is N/A for a buffer/inverter). The VSS-to-input diodes work the same way but have good hFE, so essentially all the negative ESD current is drawn from VDD.
Anyway, this is easily avoided, by adding external schottky diodes (BAT54C seems a likely choice), or using a device with an input structure that tolerates higher voltages (like the CD4504 pictured).
I don't think discrete logic has any problems with #1, so that should be safe. It is something to be aware of, in general. For instance, FPGAs can have these limitations. Many are made with input structures as small as possible -- to offer as much speed as possible, given the wide range of output types they need to be programmable for. Consequently, they can't handle much DC current. This is a reliability problem, driven by electromigration. Some specify transient voltage or current limits, others don't. (Offhand, Altera's Cyclone III was rated for VCC + 1V or VDD - 1V, for 2ns, or something like that, and 2mA DC.)
Tim