Over the last couple of months I have been versy active in several L21 deep ancestry projects and forums and here is some conclusions that may help (not for the faint of heart):
1) Almost always test for Y-STRs first. About half of the time, knowing your 67 STRs, you can estimate your SNP by looking at Mike Walsh's huge spreadsheet in the Y21 Yahoo Group. This is not for the faint of heart (or can post your Y-Search ID at 67 markers and ask for help). 37 markers is usually not enough. They are now defining new DNA fingerprints that require 111 markers and increasing the number of DNA fingerprints of various deep ancestry branches.
2) The FTDNA deep clade is constantly adding new SNP descendants of L21. As of about two weeks ago, here is the list of downstream SNPs tested:
M222, L96, L144, L159.2, L226, P314.2, P66, L193 and M37(?).
M37 has been found not to be a branch as only one person has tested positive to date. This list is growing constantly.
3) There are many SNPs in the process of being added under L21 at this point in time and can be ordered separatelynow if you are eager. DF21 is very broad branch and is expected to be the second largest branch of L21 - behind M222. DF25 is known to be new son of DF21 and DF5 is a son of DF21. L371 is another new son of L21 but is a relatively small branch which can be predicted by three 67 marker STR values. There are others on the verge of being identified as branches. All the above SNPs will be added to the ISOGG, YCC and FTDNA haplotrees officially in the next month or so and will be added to FTDNA's deep clade tests several months later. The above SNPs can be ordered now individually for $29 each.
4) There are another 15 or 20 SNPs under L21 under investigation. Around half of these (if not more) are labeled as "private" SNPs. This means that they are SNPs in the genealogical time frame (50 to 500 years old). These SNPs will never be added to the haplotree - but some SNPs can be more recent than the MRCA of your surname cluster and can divide your surname cluster into two branches. Eventually, there could be dozens of these private SNPs just under L21 - it is not really known the limit to how many private SNPs there are out there.
5) Comparing the MRCA of your most recent SNP to the MRCA of your surname cluster produces a DNA fingerprint for genealogical usage. You can use your surname cluster DNA fingerprint as input to evaluating possible NPEs. It can be used to help define your surname MRCA if some MRCA marker values are not clear (no clear majority of values for some markers). These DNA fingerprints are a much better criteria for selecting related lines than mutational difference.
6) If you are confident that you have defined a solid surname cluster and have 20 submissions, you only really need two or three tested for the SNP that defines your branch. If they are all really related at 600 years (STRs), they will obviously be related at 1,200 years (SNPs). If you are unsure about some belonging to your surname cluster, testing for SNPs can divide your cluster into two or more clusters (if they do not share a common ancestor at 1,200 years - they can share an ancestor at 600 years).
7) If you really want to be bleeding edge - you can order a partial Y-chro test for $750 from FTDNA. You have to make an application and takes several months. Around half tested do not discover any new SNPs and half discover new SNPs. This is how SNPs are found (not just from FTDNA but from several other sources as well). As more haplotree branches are found, the ratio of private SNPs to haplotree branches will constantly rise. The FTDNA test is called "Walk the Y" I am raising funds for WTY test under L226 (third or fourth largest branch under L21). I am really amazed how much deep ancestry research can benefit genealogists - it has really opened my eyes to better analysis but greatly complicates the analysis as well.