Is there a way to work out a bell curve on selective national competitions, so instead of including 0%ile to >99.9%ile, it's a bell curve across 70%ile to >99.9%ile?

Yes, this could be done, in several steps, something like this:

1- Collect the scores (whether current scores, and/or complete history of scores, or past x years of scores, etc).
2- Determine relative performance percentiles for each score and/or range of scores.
3- Identify the cut score for the 70th percentile.
4- Graph the scores which are above the cut score.
5- Draw a line to connect the actual graphed data points. The distribution may form a "normal curve" (symmetrical bell curve) or may be somewhat bumpy depending upon the data's sample size, actual scores, and frequency with which each score occurs.
6 - Draw a line to smooth out the curve, showing it to be as close to a "normal curve" (symmetrical bell curve) as possible.

Does this high-level process description answer your question???

I think if you are measuring something with an underlying normal distribution in the population (like IQ), you would not get a normal curve just focusing on the >70th percentile unless you were having a very select group of test-takers sit for it (like people who had previously scored in the 80s). In most cases I think you would still see a skewed and not normal distribution.