Added comment with reasoning for number or array BN instantiation.

lib/encoding/bufferreader.js

@@ -94,6 +94,12 @@ BufferReader.prototype.readUInt64LEBN = function() {
   var second = this.buf.readUInt32LE(this.pos);
   var first = this.buf.readUInt32LE(this.pos + 4);
   var combined = (first * 0x100000000) + second;
+  // Instantiating an instance of BN with a number is faster than with an
+  // array or string. However, the maximum safe number for a double precision
+  // floating point is 2 ^ 52 - 1 (0x1fffffffffffff), thus we can safely use
+  // non-floating point numbers less than this amount (52 bits). And in the case
+  // that the number is larger, we can instatiate an instance of BN by passing
+  // an array from the buffer (slower) and specifying the endianness.
   var bn;
   if (combined <= 0x1fffffffffffff) {
     bn = new BN(combined);