MAGNEW,TXYZ:  0.31640638004497
 U,V,W -6.56290015166051E-04  9.07632797085983E-03  0.56242650150838
 

In option MGNFIELD, one inputs the components of the magnetic field, but if
they are set all = 0, a user subroutine MAGFLD is called, with the following
arguments: ( X, Y, Z, BTX, BTY, BTZ, B, NREG, IDISC ). 
Here, BTX, BTY and BTZ ARE NOT the components, but the direction cosines!
The absolute value is given by B. Looking at the direction cosines printed
in the error message, it is clear that U, V and W (the direction cosines)
are not properly normalised. The sum of their squares is 0.316406, as shown
by the TXYZ value).

Probably you did excessive biasing of the interaction length (WHAT(2) of
option LAM-BIAS), and you got too many interactions. Try increasing the 
absolute value of WHAT(2): the smaller the reduction factor, the stronger
the biasing.

This message indicates that insufficient memory has been allocated for
the "contiguity list" (list of zones contiguous to each zone). This is
not an actual error, but it is highly preferable that the user optimise
computer time by increasing the values of the NAZ variable in the
geometry region specifications.
In particular, for each region, FLUKA sets up a list of "neighbour
regions" to be tested first when a particle leaves any of the
bodies making up that region.
If the new region is not found in the list, the other regions
are searched. When found, the region is added to the list of
neighbours, provided there is enough space left in the array.
The space allocated is by default 5 neighbours per region, which is
normally more than sufficient. The value can be modified by the user
as explained in the manual (description of region data)

  => the integer in columns 6-10 is the number of regions which can be
     entered by a particle leaving any of the bodies defined for the
     region being described (leave blank in continuation cards). This
     number is used to allocate memory, and it is not essential that it
     be exact (if left blank in number-based geometry inputs, it is set to 5).
     Any number is accepted, and only the sum of all numbers indicated for
     each region matters. There is no need for this number to be close
     to the actual value for each individual region, provided the total
     is large enough. For example, setting NAZ to 91 for 1 region and to
     1 for 9 other regions is exactly the same as setting it to 10 for
     all ten regions.

Note that memory allocation is done globally, not region by region.
In general, the impact on the CPU time can vary wildly depending on whether
the contiguity list is marginally small or severely underestimated, and on
the overall complexity of the geometry. For complex geometries, with
severely underestimated NAZ's, the impact can be huge, while in other
cases it can be hard to appreciate.

Don't worry: the "missing energy" means nothing wrong, it is a honest
physical quantity! It is what we call "Q" in nuclear physics.
When you have an endoenergetic nuclear reaction, for instance a
(n,2n) reaction, you have a positive missing energy (Q<0). It is missing
because which it has been transformed into mass of the final
nucleus. When you have an exoenergetic nuclear reaction, for instance a
(n,gamma) or a thermal fission, you have a negative "missing energy" (Q>0).
That means, energy is not missing at all, but is created out of nuclear
binding energy balance. In the end, FLUKA does its total energy balance, which
can be positive or negative. In a pure electromagnetic run, the missing
energy is practically zero. In a run with nuclear reactions, it can be
positive or negative. With a thermal neutron source or with fissionable
materials the missing energy can usually have a very large negative value.
With Pb and Ta, as in your case, a lot of energy is needed to break
the nucleus (think of all reactions having an energy threshold),
and the missing energy can be positive. The value you found of 6.3%
is not particularly high.

  Geofar: Particle in region 3 (cell # 0) in position 1.000000000E+00
  0.000000000E+00 1.000000000E+00
   is now causing trouble, requesting a step of 6.258867675E-07 cm
   to direction -2.285059979E-01 -9.412338141E-01 2.487245789E-01, error count: 0
   [...skipped...]
   Particle index 3 total energy 5.189748600E-04 GeV Nsurf 0
  We succeeded in saving the particle: current region is n. 2 (cell # 0)
  

The message indicates a real problem if repeated more than a few times.
As it can be seen, the program has some difficulty to track a particle in
a certain direction, and it tries to fix the problem by "nudging" the
particle by a small amount, in case the problem is due to a rounding
error near a boundary. If the message appears often, it is recommended to
run the geometry debugger centering around the position reported in order
to find if there is an error in the geometry description.