Burrows, Wheeler and Back

Pyth, 29 bytes

?tthu+VzSGzz}dzseMS.>L+z\ hlz

Demonstration. Test harness.

This straightforwardly divides into an encoding and decoding segment, with the code deciding which one to use with a ternary.

?tthu+VzSGzz}dzseMS.>L+z\ hlz
                                 Implicit:
                                 z = input()
                                 d = ' '

?           }dz                  If there is a space in the input,
    u     zz                     Update G to the result of the following,
                                 with G starting as z and repeating len(z) times:
     +V                          The vectorized sum of
       zSG                       z and sorted(G)
   h                             Take the first such result, which will consist of
                                 the first character of z followed by the
                                 first cyclic permuation of the pre-BWT string,
                                 which must start with ' '.
 tt                              Remove the first two characters and return.
                 L               Otherwise, left-map (map with the variable as the 
                                 left parameter and a constant as the right)
               .>                cyclic right shift
                  +z\            of z + ' '
                      hlz        over range(len(z)+1)
              S                  Sort the shifted strings,
            eM                   take their last charactes,
           s                     combine into one string and return.

CJam, 41 36 35 bytes

q:XS&LX,{X\.+$}*0=1>XS+_,,\fm<$zW>?

Test it here.

Explanation

q:X   e# Read STDIN and store it in X.
S&    e# Take the set intersection with " ". We'll use this as a truthy/falsy value to
      e# select the correct output later.

# Compute the iBWT:
LX,   e# Push an empty array, compute the length of X.
{     e# Run the following block that many times:
  X\  e# Push X and pull the other array on top.
  .+  e# Add the characters of X to the corresponding line of the other array,
      e# i.e. prepend X as a new column.
  $   e# Sort the rows.
}*
0=    e# Since we just sorted the rows, the first permutation of the output will be
      e# one starting with a space, followed by the string we actually want. So just
      e# pick the first permutation.
1>    e# Remove the leading space.

# Compute the BWT:
XS+   e# Push X and append a space.
_,    e# Get that string's length N.
,\    e# Turn it into a range [0 .. N-1], swap it with the string.
fm<   e# Map each value in the range to the string shifted left by that many characters.
$     e# Sort the permutations.
zW>   e# Transpose the grid and discard all lines but the last.

?     e# Choose between the iBWT and the BWT based on whether the input had a space.

Perl 5, 179

Another not-so-good one from me. There are probably some gains to be had in this one, but it's not going to compete with the purpose-built golf languages.

$_=<>;@y=/./g;if(/ /){@x=sort@y;for(1..$#y){@x=sort map$y[$_].$x[$_],0..$#x}
say$x[0]=~/^ (.*)/}else{push@y," ";say map/(.)$/,sort map{$i=$_;join"",
map$y[($_+$i)%@y],0..$#y}0..$#y}

Un-golfed:

# Read input
$_ = <>;
# Get all the chars of the input
my @chars = /./g;

if (/ /) {
  # If there's a space, run the IBWT:
  # Make the first column of the table
  my @working = sort @chars;

  # For each remaining character
  for (1 .. $#chars) {
    # Add the input as a new column to the left of @working,
    # then sort @working again
    @working = sort map {
      $chars[$_] . $working[$_]
    } 0 .. $#working;
  }
  # Print the first element of @working (the one beginning with space), sans space
  say $working[0] =~ /^ (.*)/;
} else {
  # BWT
  # Add a space to the end of the string
  push @chars, " ";
  # Get all the rotations of the string and sort them
  @rows = sort map {
    my $offset = $_;
    join "", map {
      $chars[($_ + $offset) % @chars]
    } 0 .. $#chars
  } 0 .. $#chars;

  # Print all the last characters
  say map /(.)$/, @rows;
}