I recall asking a theoretical physicist the same question at a similar
stage in my career; the answer was 'do a degree in either maths
or physics at the best possible uni'. I am a theoretical physicist these
days, and the advice still seems about right.
For more detail, maybe it's best to work backwards from your goal,
which is really to get a PhD place with the best possible supervisor
(and choosing him/her is a whole other story!) while having the
appropriate knowledge.
On both these counts you really need to do the course content
of an MSc in Particles and Fields, or similar. IC, Durham and King's
London run them; perhaps a few others. The classic equivalent is
to do the Cambridge CASM ('Part 3', the 4th year of the maths course).
I'm quite surprised that a friend of yours from NatSci Physics was not
allowed to do this; it's a classic route into the field.
To get there you could do either NatSci or Maths at Cambridge. There
are plenty of top theoretical physicists who've come through each route,
and many from equivalent degrees elsewhere. (My impression is that,
among the very best, the Physicists perhaps shade it over the Mathematicians,
of whom I am one. The benefits of a good education in Physics are perhaps
slightly greater than those in Maths, much of which one can teach oneself later.)
If you're not at Cambridge then the key things are
1. be at a uni where there's a continuum of research interests
across the maths/physics divide. This usually means a uni with
strong theoretical physics in the maths depts. Durham is the UK's
leading example - and several prominent academics started their
careers there. You can tell by looking at staff research interests
and the range of final year modules and projects available.
Don't go somewhere that really has no modern, fundamental
theoretical physics research. You can easily identify these.
2. be among the best possible peers. Other things being equal,
the unis with the strongest students can teach the most advanced
material. Look for unis where the last year of any four year course
you choose includes modules in QFT and GR. These should be the
core of your final year: other physics (cosmology, condensed matter,
astroparticle stuff,... ) and maths (diff geom, Lie algebras, ...) are
less crucial. That will mean that you can cover about half of the
Pt3 year. Less than this and you may be lagging too far behind;
getting on one of the aforementioned MScs thereafter then becomes
your only realistic way forward.
3. get a first class degree, and preferably be among the top few
of your year. This will help you get the best PhD supervisor. To be
frank, if you don't reach this level, you'll have a very great deal
to prove in your PhD (actually everyone has, anyway), and you'll
be a remarkable person indeed if you do so! (There are a few -
people who were deep thinkers but who perhaps didn't have the
speed for exams.)
Hope this helps.