Here is some example code that demonstrates the penalty incurred with creating a NSDateFormatter every time you need to format a date instead of creating it once and storing it for use when needed.

FormatterTest.swift

import Foundation
import QuartzCore

public func testWithMultipleInstantiation() -> CFTimeInterval {
    var dateStrings: [String] = []
    dateStrings.reserveCapacity(100000)
    let start = CACurrentMediaTime()
    for _ in 0..<100000 {
        let df = NSDateFormatter()
        df.dateStyle = .MediumStyle
        df.timeStyle = .MediumStyle
        dateStrings.append(df.stringFromDate(NSDate()))
    }
    let end = CACurrentMediaTime()
    return end - start
}

public func testWithSingleInstantiation() -> CFTimeInterval {
    var dateStrings: [String] = []
    dateStrings.reserveCapacity(100000)
    let start = CACurrentMediaTime()
    let df = NSDateFormatter()
    df.dateStyle = .MediumStyle
    df.timeStyle = .MediumStyle
    for _ in 0..<100000 {
        dateStrings.append(df.stringFromDate(NSDate()))
    }
    let end = CACurrentMediaTime()
    return end - start
}

public func testWithStaticMethod() -> CFTimeInterval {
    var dateStrings: [String] = []
    dateStrings.reserveCapacity(100000)
    let start = CACurrentMediaTime()
    for _ in 0..<100000 {
        dateStrings.append(NSDateFormatter.localizedStringFromDate(NSDate(), dateStyle: .MediumStyle, timeStyle: .MediumStyle))
    }
    let end = CACurrentMediaTime()
    return end - start
}

Playground.swift

import Foundation

let timeForMultiple = testWithMultipleInstantiation()
let timeForSingle = testWithSingleInstantiation()
let timeForStatic = testWithStaticMethod()

print("multiple:", timeForMultiple) //multiple: 7.96959602891002
print("single:  ", timeForSingle)   //single:   0.888885863008909
print("static:  ", timeForStatic)   //static:   7.89041009696666