BigDecimal.subtract(BigDecimal subtrahend) is currently implemented as:

    public BigDecimal subtract(BigDecimal subtrahend) {
        return add(subtrahend.negate());
    }

subtrahend.negate() produces a new BigDecimal object since BigDecimal objects are immutable.

This object allocation could be avoided by providing a "pure" implementation of subtract, similar to the "pure" implementation of BigDecimal.add(BigDecimal augend).

A prototype of a "pure" BigDecimal.subtract(BigDecimal subtrahend) suggests an improvement of 1% - 2% in SPECjbb2005 performance on some hardware platforms.