I think tapes in proper data centres are reliable. At least for several years. Drives are monitored, error rates checked, and the configuration is stable.
However, tapes on ad-hoq systems have many problems. One significant problem is the compatibility of the software to read the tapes. Companies keep evolving their software, and only have compatibilty for a few previous versions. SCSI configuration can be difficult, and tapes can be written with errors due to SCSI, cables etc, and speed related issues.
To keep data for a long time, it has to be re-processed every few years, and always kept on duplicate tapes.
For most users, disk drives are easier, and although capacities change, the basic operation is much more stable. We've had FAT for 30(?) years and NTFS for 20(?) years.
Personally I would suggest that now disk/RAID is the best solution, but tape can be made to work if used very carefully.
Let's be accurate about those two numbers - 10-50% is a big swing, and 10-50% plus 77% does not make a greater than 100% chance of failure.
Well, to be accurate, and without the original data, it seems to me that the two things cannot be "summed up".
The 10-50% rate of failures is a (BTW "vague" enough range) quantity relative to attempts of restoring and their failure rate.
The 77% is seemingly ANOTHER thing, it is the amount (among those that REPLIED) of people that EVER experienced a failure.
Or, if you prefer, if you ask 100 people in the business, and they ALL reply, 23% will telly ou that they NEVER experienced a backup on tape restore failure. 😯
The sheer fact that the latter data comes from a (I presume voluntary) survey makes the relevance of the data actually scarce from a statistical point of view, there is clearly a bias in it, in the sense that if you ask about failures in a survey, it is more likely that people that actually experienced that failure will respond, but anyway even if the result is accurate it is completely pointless, without a "weight".
Let's take the same 100 people sample, if (purely speculative data)
77 reply Yes, I got at least one failure, and overall I made 100 attempts
23 reply No, I never got a single failure and overall I made 100 attempts
the result is that out of 100*100=10,000 attempts, 77 failed, and that would make the method have a failure rate of 0,77%.
To get the 10% you need
77 I had 13 failures out of 100
23 I had none out of 100
77*13=1,001/10,000= 10%
To get the 50% you need
77 I had 65 failures out of 100 ?
23 I had none out of 100
77*65=5,005/10,000= 50%
jaclaz