All drivers are broken in at high signal levels by 25hz sine wave non-stop for two days. Then they are measured very carefully using delta compliance with LspLab and finally matched in pairs based on their Thiel & Small parameters.   With this procedure I can guarantee the technical specs of the speakers for a long time since they are broken in and, hence, do not change when used by the customer.  

Thus far I have had little problem with 3% matching.  Most of the time I am able to get them within 2%.  This, however, is not the primary reason that I match the drivers.   While considerable 10% variation is common and audible, I don't measure drivers because matching drivers makes a "huge sonic improvement".  I match drivers for quality control.  I have encountered bad drivers - too often.

A bad driver will measure bad and sound bad.  I have measured such drivers, and heard the results.  I have received them in kits.  I have also received email from others who received them in kits.  Implementing these drivers results in bad sound.  It also results in bad business.  It is just "BAD" from every angle.  It would seem that the "better" drivers would always be perfectly operational direct from the factory.  Such is not the case.  I once purchased eight Vifa wide range drivers.  Four of them had scratchy voice coils -  bad.  I have measured 100-200 drivers (not many),  and have found two with a perfect record of Quality Control.  One is the SCC 300 from Jeff Glowacki.  The other is the Hiquphon tweeter from Oskar Wroending. 

I generally do not measure crossover components.  After measuring hundreds of components I find little reason for this.  Even the mid-range price components (i.e. Solen Capacitors and Madisound Inductors) are extremely consistent.  I don't remember a variance of greater than 2%.  Quite often the variance is less than 1%.   This is normal consistency for crossover components.  This is miniscule when compared to driver variance that always occurs.  Also, it might sound nifty to have capacitors matched to within 1%, but this is not audible.  Matching crossover components to anything less than 1% exists solely for the marketing department or those who are uselessly anal.  I will admit to this extreme detail orientation at times, but please don't mistake my comments regarding THIS matter for something significant in the audible realm.  The only significant deviation I have found is that some inductors will consistently measure on the low side of their advertised tolerance, but they remain very consistent.