EN50163 actually has different frequency limits for systems with no synchronous connection to an interconnected system (e.g. supply systems on certain islands). These have much wider tolerances, +/-2% (49Hz - 51Hz) for 95% of a week and +/-15% (42.5Hz - 57.5Hz) for 100% of the time. I think this reflects the increased difficulty of maintaining a stable frequency in a small network of synchronous generators with rapid load variations.
(There are also three different sets of frequency limits for 16.7Hz systems, but I think that is OT for this thread).
To me the wording of the Frequency section of the EN bears the hallmarks of a typical standardisation committee compromise/fudge between the opposing views of different interest groups. The supply industry, in their CENELEC committee, would have insisted on keeping 47Hz as the extreme limit for the grid. But the train makers, in the railway committee, would have argued that 47Hz was unrealistically conservative and would unnecessarily penalise the design of transformers and traction converters. So they got the note added to say that they only really had to design for 49Hz.
The chickens are now coming home to roost. Wind and solar farms, and the continental interconnectors, use induction generators or inverters that just follow the frequency of the reduced number of synchronous generators in thermal and nuclear power stations. So there is less spinning mechanical inertia to smooth out short term frequency fluctuations when there is a sudden change in the balance between supply and load. At the same time train transformers and traction electronics have become more finely tuned to the supply frequency, to maximise efficiency and minimise weight, and so are less tolerant of fluctuations.