Yes - it can lead to spurious conclusions - howver, when done properly it can cetainly benefit.
E.g.g A lot of testers might try to select a system based on data up to,say, 5 years ago- and then select their method for the next 5 years to see does it stand up.
Also - other testers may purposely NOT select the best performing method in the belief that the same trends will not occur in the future.
Many testers will only use the most robust method i.e. the system that stands up best to a variety of ALL assets - as opposed to applying a method for buying say,gold in the future that historically worked very for well specifically for gold in the past.
My point is - backtesting is far from a holy grail,but,when applied correctly, it is certainly better to have this information than not have it.