It looks like the LPC11xx series of parts have a timer that can accept an external crystal. Wire up a 32kHz crystal, set the timer to divide by 128 and have the timer interrupt. I would be very surprised if you could not wake from deep sleep with this mechanism.
Before getting that far though, you mention that the LPC11xx has a very inaccurate wake from deep sleep. It sounds like no matter what method you use to wake it from deep sleep, it is the wake from deep sleep that is the problem. What are the bounds on this wakeup? What if you divide by 64 instead, and only doze for the remaining time? You'd still spend a lot of time in deep sleep, but your "snooze" function would let you get that accurate sample rate.
[1.9ms deep sleep] [wake from deep sleep] [1.9ms - wake time doze] sample
If you're using a timer with an external crystal you could very accurately time that third part (1.9ms - wake time).
Honestly though, it sounds like you might be using the wrong chip for this task. I've not used them myself, but I've heard very good things about the MSP430's power consumption. You might be able to achieve better power consumption and get the ADC results you want with a part more suited for extreme low power sampling.