1. Prescription to a Isodose : This essentially means that you have a given dose prescribed at a given isodose. The way of choosing this was arbitrary but in my early IMRT days where we were worried about failures we normalized to a percent isodose that covered more than 99.5% (we were looking for 100% though). That meant that if the 95% isodose covers the target we would prescribe to that.
2. No Plan normalization: Allowing IMRT to do the leg work.

Is there any data on this. I could find a paper in medical dosimetry here
The essence of the conclusion is that you can have unexpected high dose or low dose with plan normalization to a percent isodose. To quote the final conclusion of the paper:

….these normalizations help meet the volume coverage goals, but unlike 3D planning situations, may result in avoidable delivery of additional doses to the normal tissues. The focus of this study is to evaluate the effect of application of normalization for IMRT planning using multiple patient situations. Recommendations would favor re-optimization over normalization in most planning situations.

In have seen this situation crop up in SIB IMRT particularly where there was the dilemma of which target volume you select .. usually plan normalization to say 97% for the HRCTV will not imply the same coverage for LRCTV. The OAR doses also tend to increase and that increase needs to be looked after too which I think the authors highlighted in the paper. Having seen the optimization done here at Calgary (i.e. no plan optimization) - I am inclined towards that solution .. if you feel the coverage is not coming reoptimize and get the oncologist involved .. usually the target volume may not have been drawn correctly. Often changing the constraints according to the input of the oncologist and physicist can get a better plan ready.
Optimization to mean and median ? - I guess the result would mean a higher or lower dose depending on where the mean and median is. Then again the question comes which mean and median you choose if you are doing SIB IMRT. Even in simple IMRT the mean or median being measures of central tendency would tend to be affected by the extremes in the dose distribution (particularly for mean). Although it will standardize the plan normalization method it will not standardize the dose distribution I think. Particularly for mean I have seen that it can really show some strange values and the median dose is usually higher than the prescribed dose .. so you would tend to give lesser dose to the PTV if you prescribed to the median than you would if you had no plan normalization or normalization to a isodose.
That said I am not a physicist and input from some of the physicists is welcome here.

Commercial Treatment Planning Systems contain various plan normalization options

e.g

1. No Plan Normalization
2. Plan Normalization at Isocenter
3. Plan Normalization at Target Mean
4. Plan normalization at Target Minimum
5. Plan normalization at Target Maximum
6. % of Volume should Receive _ % of dose
7. Etc.,

Normalization at Isocenter is no long applicable to IMRT

Most of the institute uses no plan normalization. But the pitfall of this normalization method is it may be under dose / over dose to target if the constraints are not properly handled by the planner. So, it is purely depends on the planner constraints & priorities.

I personally would like to use plan "normalization at target mean / Median", which can standardize the plan normalization method in IMRT and it is “independent” of the Planner / user dose constraints & priorities.

I will appreciate if the forum further discuss about plan normalization in IMRT and their institute protocols of IMRT Plan normalization.